OR in the Press – Page 5 – Mike Trick's Operations Research Blog

Business Intelligence and Operations Research

In the past couple of years, a field called “business intelligence” has sprung up. Based on the premise that businesses should get more out of data, business intelligence mixes data mining, algorithms, visualization and other approaches to help businesses make better decisions.

Of course, I thought that was the definition of operations research! Ever since I came across the area, I have included some of the blogs in my blogroll (see, for instance, James Taylor’s Decision Management and Smart (Enough) Systems). I find this area interesting, but I never can quite get my brain around what they are trying to say. It is like they are taking an area I know very well and translating it into a different language which I kind-of understand, but not quite well enough to grasp what they are saying.

Intelligent Enterprise (part of the group that publishes Information Week) has a short article “What BI Practitioners Can Learn from Operations Research”. It begins with the Netherlands Railway Edelman story, then continues to express confusion on the lack of interaction of the two fields:

It would be natural for BI practitioners to embrace OR, which has long focused on automating decision making, surely the goal of those who talk about closed-loop BI. “OR starts with the decision and works back to figuring out what math and data will help with devising a better solution, while BI tends to start with the data and see what can be done with it,” says James Taylor, co-author of Smart (Enough) Systems and one who believes that OR and BI are complementary but quite different. “OR folks tend to be focused on the nitty-gritty of day-to-day operations, and they use data from operational systems. BI tends to be focused on knowledge workers, data warehouses, and aggregation.”

It would be natural for the OR community to reach out to the BI world and its community of business-focused knowledge workers, who are increasingly looking to build out their analytical toolkits. “C-level decision makers are turning to analytics for help in the decision-making process,” writes Peter Horner, editor of Analytics, a new magazine published by the Institute for Operations Research and the Management Sciences (INFORMS). “When you see terms like operations research (OR), think analytics.” Many in the BI world, who are already supporting those executive decision makers, are saying close to the same things about BI and analytics.

Given the close kinship of BI and OR, one wonders why these two camps have long existed as separate communities?

SAS’s Mary Crissy (who I still think of as Major Crissy, though she left the military some years ago), has what I think is a pretty good explanation:

“Operations researchers don’t interact with the IT community as much as they ought to,” says Mary Crissey, an analytics marketing manager at SAS, a council officer of INFORMS, and, apparently, one of the few vendor executives with a foot in both the BI and OR camps.

“Academic mathematicians are not worried about what terms are buzzing about in the business world,” Crissey says. “They talk to each other in their mathematical language of equations and theory without getting entangled in terminology such as BI. Pure Intelligence for business or public service organizations all boils down to data analysis; they just don’t call it BI.”

Having gone through fights over “operations research”, “management science”, “decision engineering”, “analytical decision making” and countless others over the 50+ years of existence, the field is not particularly excited about embracing a new name for our field.

I guess I see BI’s relationship with OR to be similar to operations management’s relationship with our field. OM uses OR an awful lot, and OM would not be successful as a field without OR. But OM is not a subfield of OR: sometimes it uses approaches that are outside the range of OR (including organizational theory, case studies, or other methods). That is great! OM people are trying to solve problems, and they should be using whatever methods seem appropriate. Similarly, BI uses (or should use) OR extensively. And OR people should see the BI community as a great source of problems and inspiration (and should make an effort to learn their language). But BI will inevitably use non-OR methods for some of their issues, so is rightly not “the same as” OR. But we as a field should know more about what they are doing if we are going to be part of this business direction.

Ant Colonies in the Skies

“Discoveries and Breakthroughs in Science” is a producer of short TV clips on results in math and science. INFORMS is involved with them, and is seeking story ideas to pitch to them.

This month’s mathematics story is about using ant colony optimization to help run an airport, with an emphasis on the gate assignment problem.

“It’s sort of like a colony of individuals trying to move through a maze with all of the other individuals present, arriving and departing and trying to do it as fast as they can,” Douglas Lawson, Ph.D., a financial analysis manager at Southwest Airlines in Dallas, Texas, told Ivanhoe.

The software program uses swarm theory, or swarm intelligence — the idea that a colony of ants works better than one alone. Each pilot acts like an ant searching for the best airport gate. “The pilot learns from his experience what’s the best for him, and it turns out that that’s the best solution for the airline,” Dr. Lawson explains.

I’m not a huge fan of Ant Colony Optimization: it has always seemed overblown to me. But the proof is in the results, so if this does work better than alternative approaches, then that’s great. I have not found Lawson’s work on this on the web, but gate assignment has been tried with ACO (see here and here) and other forms of swarm intelligence, and Lawson is the “Manager of Process, Forecasting, and Simulations” at Southwest, an admirable airline, so it is interesting to see him using this in practice.

One thing bothered me and one thing I found humorous. First, Lawson and his team are described as “financial analysts”. Is the “Manager of Process, Forecasting and Simulations” a financial analyst? My guess is that the TV producers decided not to make the big step and call him “operations researcher” so they went with a word they thought the audience would know.

The humorous part comes in the background information where they describe swarm intelligence:

HOW DO SWARMS OPERATE? How do ants find a route to a food source? Each ant follows the strongest pheromone (chemical) trail left by other ants. If this process is repeated frequently enough, they will find the best route through trial and error. If ants become isolated from their group, they end up running around in circles, following their own pheromone trail until they die of exhaustion. This behavior, called “swarm intelligence,” …

I would hate to be on the plane that gets separated from the others!

Green Supply Chains

Environmental modeling has been an increasingly active area of OR over the past few years (see the greenOR blog for many examples). As companies strive to do good, either for economic or other reasons, they are thinking more about environmental impact in all that they do.

ILOG has just announced a “Carbon Footprint” extension to its supply chain modeling software (which in turn is based on the software from David Simchi-Levi’s previous company LogicTools). InfoWorld is one site that has picked up on this:

In an effort to aid companies as they struggle to balance profitability and environmental responsibility, vendors are rolling out increasingly sophisticated tools. Among those vendors is ILOG, which this week released a Carbon Footprint extension to its LogicNet Plus XE supply-chain application. This remarkable tool serves a valuable function: It’s designed to help companies evaluate the impact that various supply-chain network configurations and transportation strategies would have on their carbon footprint.

David Simchi-Levi is quoted regarding a case of a company trying to decide how many distribution centers to have:

Using LogicNet Plus XE with the Carbon Extension, the company cranked out various scenarios that involved adding between two and seven new distribution centers. Turns out that moving to four distribution centers would have resulted in the highest costs savings. Thus, a company that wasn’t thinking about green metrics might have gone with that option.

The company found, however, that by going up to six distribution centers, it would have slightly higher costs (1.6 percent), but it would reduce transportation distances by 20 percent and overall carbon emissions by 11 percent.

Those results might come as a surprise: How could adding six more energy-consuming distribution centers result in less carbon waste? The answer: With the six-center model, the company relies more on trains for transporting goods inbound than it does on trucks to ship products outbound. Trucks have a significantly higher environmental impact than trains, according to Simchi-Levi.

I don’t know whether a 1.6% cost increase is worth an 11 percent reduction in carbon emissions, but having the information makes it possible to make an informed decision.

This reminds me of work being done in robust optimization, of various forms. David Ryan of the University of Auckland spoke two years about his efforts to devise airline schedules that are less susceptible to delays (I wrote on this in 2006). He found that a very small cost increase allowed a huge improvement in the robustness of the schedules. Many of these supply chain optimization problems are relatively “flat” with many near-optimal solutions (cases where there is only one optimal solution with everything much worse tend to have obvious solutions). In such a situation, “secondary” issues such as environmental impact, customer perception, or robustness to variation in data take on a higher importance.

OR at P&G

Let me be late in the OR blogging game and note that there is a great article on operations research at Proctor and Gamble on bnet. It is wonderful advertising for our field, including phrases like “P&G’s Killer Apps in OR”. INFORMS was strongly involved in the article, with quotes from Past-President Brenda Dietrich and Executive Director Mark Doherty:

P&G, GE, Merrill Lynch, UPS — the list of Fortune 500 companies getting into the OR game is expanding, says Mark Doherty, executive director of the Hanover, MD-based Institute for Operations Research and Management Sciences (INFORMS), an OR think tank. “In the private sector, OR is the secret weapon that helps companies tackle complex problems in manufacturing, supply chain management, health care, and transportation,” he says. “In government, OR helps the military create and evaluate strategies. It also helps the Department of Homeland Security develop models of terrorist threats. That’s why OR is increasingly referred to as the ‘science of better.’”

Having sat in on a few too many board meetings, I think calling INFORMS an “OR think tank” is going a little far. But the article does project the very best vision of our field.

Check out another take on the article at Punk Rock Operations Research (and I thought I saw it on another OR blog, but it escapes me at the moment).

Operations Research to Decide the Election?

Brian Borchers wrote me to comment on an article in the New York Times on how the US primaries are moving into a phase characterized by complicated resource allocation problems. Up until now, it was easy: candidates could spend their time in Iowa and New Hampshire (then Nevada and South Carolina) and not feel overstretched. But with those primaries and caucuses now past, it gets more complicated. Tuesday February 5 is “Super Tuesday” when 24 states are to choose their delegates to the national convention. No candidate can reasonably campaign in all parts in all of these states. But the rules on how delegates are chose make it necessary:

…the delegate rules for Democrats and for Republicans are different and, within each party, often vary from state to state. For example, the Republicans have some states where the statewide winner gets all the delegates, providing an obvious target for a candidate who might seem strong there. Among them are Missouri, New Jersey, New York and Utah.

But there are other states where the delegates are allocated by Congressional district, sometimes winner-take-all, and sometimes proportionally.

By contrast, Democrats eliminated the so-called winner-take-all rules. Instead, delegates are allocated depending on the percentage of vote each candidate gets in a Congressional district, under very expansive rules that, generally speaking, mean the candidates divide the trove evenly assuming they get more than 30 percent of the vote. There are also some delegates allocated statewide, again proportionately.

As Brian summarizes:

The various political campaigns have been building statistical models to predict voting outcomes in different congressional districts and using these as the basis for game theoretic decisions about how best to spend their limited funds and limited candidate time. The more traditional polling approach isn’t adequate, because it would be too expensive to do separate polls for every congressional district…

This is an interesting game involving things like parity (if there are 2 delegates for a district, then it is enough to get 34% of the vote to get 1; with 3 candidates, 51% can earn you 2 delegates), resource allocation, timing (Rudy Giuliani chose to skip the initial rounds to concentrate on Florida and the Super Tuesday states, perhaps losing too much “momentum” in the process) and so on. Since it has been a while since both the Democratic and Republican races have been this open, this should spur interest in this sort of modeling. Punk Rock Operations Research has a posting on forecasting and polling, but that should only be the data for making better decisions. I would love to hear of any real operations research being used by the campaigns.

Simulation and the NHL playoffs

Growing up in Winnipeg, Canada (city motto: “At least it is a dry cold”), I had a short and rather forgettable hockey career (though getting a shutout as a goalie at 12 years old remains one of my favorite memories). I have been greatly outdone by my nephew Mathieu, who actually looks like he knows what he is doing when facing a shot. Since then, I follow hockey mainly through my local team, the Pittsburgh Penguins. I have been lucky to see a number of amazing players on the Penguins: Lemieux, Jagr, and “Sid the Kid” Crosby perhaps best of all. I keep hoping to provide the schedule for the NHL, which I think is the only thing that would impress my buddies back in Winnipeg.

Armann Ingolfsson of the University of Alberta has appeared in an article in the Toronto Sun on using simulation to predict who will make the playoffs this year. You can read a more detailed description of what he does in an article published in the INFORMS Transactions on Education in 2004. The main idea is to use simulation to generate 500 possible continuations of the regular season and to determine how often every team makes the playoffs. A critical feature of this is the ability to assign probabilities of wins for every game.

I am very happy to see that the Penguins currently have a 92% chance of making the playoffs. But the system doesn’t take into account that Crosby has a “high ankle sprain” keeping him out for two months. That might knock off a percentage point or two.

Operations Research and Fantasy Football

After a very successful year, my fantasy football teams are crashing and burning in the playoffs. For those who do not know fantasy sports, fantasy football involves a group (8-12 people) drafting NFL players at the beginning of a season. Each week, my team gets points based on the success (or lack thereof) of the players in their “real” games. If my players get more points than my opponent’s, then I win. After a regular season, the best fantasy teams in the league then face off in the playoffs. Some fantasy sports work a bit differently: most fantasy baseball leagues collect statistics from the entire year and give points in the final year standings on the categories, without the head-to-head matchups.

One of my teams this year was the best I ever had. I had the top three wide receivers in the league (Moss, Owens, and Edwards), the second best quarterback (Romo), a top running back (Addai), a great defense (Patriots), and a top kicker (Folk). My tightend was not great, but that was the only weakness. Over the regular season, I outscored the second best team in the league by an average of 30 points in a game (teams typically score 60-120 points in a game). But, sure enough, the playoffs come around, my team goes cold, and I lose the first playoff round. So now I am just struggling to get third place in the league.

I am not the best fantasy player around. To be so would require much, much more time than my three-year-old son will allow me. But I do try to use a bit of operations research thinking in my play. Some (but not all) key decisions come in the initial draft. Players are taken in turns in a serpentine fashion: if there are 10 fantasy players, then the person with the first pick will next get the 20th pick; the person with the 10th pick also gets the 11th pick). Given the projected player values (which is the real key to the problem: data!), the “best pick” depends on what you expect others to do, and the relative value of the alternatives. For instance, if you need a quarterback, and the best quarterback is worth, say, 280 points, with the next best being worth only 200 points (a huge difference), then you better pick that quarterback (unless you are absolutely sure that quarterback will be available when you pick next). But if there are five quarterbacks in the 280 range, you can afford to wait, since it is more likely that one of them will still be available when you pick next.

Mike Fry, Andrew Lundberg (both of the University of Cincinnati) and Jeffrey Ohlmann (University of Iowa) analyzed this issue in depth in an article published in the new Journal of Quantitative Analysis in Sport. Their work got a fair amount of press a few months ago, including a writeup in USA Today (sorry I missed it earlier: New Zealand doesn’t cover football well!). I just went through the article (while waiting for my son to wake up and enjoy Christmas), and it is terrific, going well beyond the obvious points. I particularly liked the analysis of the value of each of the draft positions. There is a view that drafting early is best, but the serpentine nature of the draft evens things out. With the data they looked at, it is true that the first position is best, but the differences are quite slight, and the value is not monotonic in draft position.

At the end, though it comes down to player projections. As the article quotes:

“If you value Ryan Leaf as the best quarterback, it’s going to tell you take him when it’s time to take a quarterback,” Ohlmann says. “If you give me bad projections, I’m going to give you some very bad advice.”

Crack, Cocaine and Operations Research

It might not be the most “Christmas-y” posting, but Al Blumstein of Carnegie Mellon (whose work I have discussed before) is quoted in the AP news coverage of the sentencing guidelines for crack versus powder cocaine. In particular, he talks about the violence that crack created:

When crack first became popular, there was an increase in murders and other crimes associated with the drug. But the bloodshed was not necessarily the result of something inherent in crack.

Instead, most of that violence was typical for what happens when any illegal drug is introduced and drug dealers with guns compete for new markets, said Dr. Alfred Blumstein, a professor of urban systems and operations research at Carnegie-Mellon University.

This shows the sort of clear-thinking that OR engenders. It also shows the value of having “Operations Research” in a professorial title: it is important for more of our work to be associated with that term.

On that note, my very best wishes for the holidays to all!

OR in Popular Mechanics

When I was a kid, I loved the magazine Popular Mechanics. In addition to articles on futuristic cars and planes, they always had articles on how things worked, and I seem to recall mechanically oriented projects that were always just outside my abilities. As time went on, I realized that my mechanical abilities were limited indeed, so I moved on to the more cerebral Scientific American and mathematics. These days, I am always amazed when I see Popular Mechanics in bookstores: it is like a blast from the 60s.

Blake Nicholson of the University of Michigan wrote me to point out that a recent article in the magazine has a heavy OR focus. In an article on Improving Air Travel, one of the ten possible improvements, changing boarding strategies, is explicitly an OR approach. A few of the other suggestions, including re-pricing landing slots to encourage better spreading of planes and the use of RFID in luggage tracking, are also OR approaches to air travel problems.

Ratcheted up a logarithm?

I really like the Atlantic magazine. Its articles are well-researched, in-depth, interesting, topical, far-ranging and generally a pleasure to read. It is a sign of my enthusiasm for the magazine that I pay the equivalent of US$15 to buy it here in New Zealand, and I take it to a local cafe/pub overlooking Waiheke (see my New Zealand blog for a couple hundred photos) to savor during one of the few periods I spend away from either work or the ever-fascinating and ever-challenging Alexander. I was reading a typically well-written and persuasive article by Andrew Sullivan on why Barack Obama is the candidate to transcend our political divides (see also his blog entry), when I came across the following:

The war on Islamist terror, after all, is two-pronged: a function of both hard power and soft power.

…

Consider this hypothetical. It’s November 2008. A young Pakistani Muslim is watching television and sees that this man – Barack Hussein Obama is the new face of America. In one simple image, America’s soft power has been ratcheted up not a notch, but a logarithm.

Ratcheted up a logarithm? What could that possibly mean? I have grown used to the phrasing “increased exponentially” even when the increase is quadratic (as a referee, I see this at least twice a year in professional papers; the popular press is hopeless on this). But can there be a less appropriate term than “ratcheted up logarithmically”? Here are some choices:

Instead of going up by one (“a notch”) it has gone up by the logarithm of one. Um, no… that would be going up by 0, no matter what base you go with.
Instead of increasing by one, you take the logarithm of the level. So if you start at 100, instead of going to 101, you go to the logarithm of 100. We’ll go with base 10, so that would take you to 2. Wrong way (and will be for any reasonable base)… no.
Instead of increasing by 1, you increase by the logarithm of the level. So if you start at 100, you go to 102, instead of 101 (base 10). Well, at least it has the right direction, but I can’t believe this is what Sullivan had in mind.

I do not believe I am overly picky here. If Sullivan had referred to “Congresswoman Clinton”, clearly any fact-checker, editor, or casual reader would correct/castigate as appropriate. Or an illiteracy, “Obama is the bestest choice”, would not possibly make it through the system. But the mathematical illiteracy of the population is sufficient that this can go through in what has to be one of the most closely edited magazines in existence.

This reminds me of the old question. Why is it when I say “I am in Operations Research: the science of using mathematics to make better decisions” (or whatever phrase I am going with at the moment), it is quite common for people to say “Oh, I never understood any mathematics”, expecting me to understand and applaud them for recognizing their limits. But when they refer to a book, if I were to say “Oh, I never understood that reading thing”, I would look like an idiot and they would move on to someone else to talk to. Why is there this difference?

If I am blogging 50 years from now, I am sure I will be complaining “Nice article, too bad it doesn’t mention operations research by name”. But, I would hope that I am not seeing lots of “ratcheted up by a logarithm”!